Data Engineering for Multimodal AI

Chapter 4. Ingestion and Transformation for Multimodal AI Systems

A Note for Early Release Readers

With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles.

This will be the 4th chapter of the final book. Please note that the GitHub repo will be made active later on.

If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at [e-mail].

Our industry loves bold claims: “ETL is obsolete,” “ELT is universal,” “streaming is mandatory.” Here is another one. No matter how advanced your models and AI agents are, they are constrained by the architecture that delivers their data.

In practice, ingestion is a set of trade-offs. Those trade-offs become more complex in multimodal systems, where text, images, telemetry, and events arrive asynchronously. Decisions about batch versus streaming, early versus late transformation, and validation boundaries directly shape latency, cost, reliability, and model performance.

This chapter distills the core design patterns behind multimodal ingestion. We compare batch and streaming pipelines, examine transformation strategies across modalities, and revisit ETL versus ELT through the lens of AI workloads. Using a fictional cloud kitchen company, Byte Eats, as a running example, we trace how real-time sensor data, customer reviews, and ...